Skip to content

Latest commit

 

History

History
54 lines (41 loc) · 2.77 KB

README.md

File metadata and controls

54 lines (41 loc) · 2.77 KB

RMRR DeepStack model birds

A concept DeepStack YOLOV5 model for detection of what kinds of birds are at my feeder. Note I guessed at many of these so the names might be wrong but each class should be close to being all mapped the same kind of bird.

How to use:

To use the model download the birds.pt file to your C:\DeepStack\MyModels folder. (Can be any folder but Blue Iris defaults to C:\DeepStack\MyModels)

  1. Add a folder inside your DeepStack instance (If you run on Host) or mount it in Docker.

  2. Run DeepStack with the command --MODELSTORE-DETECTION "C:/DeepStack/MyModels" See DeepStack utils for more info on setting up and testing your set up.

  3. To use RMRR detection call the end point /v1/vision/custom/birds with your picture and you get a response similar to this if it finds a known object:

{
'success': True,
'predictions': [{
  'confidence': 0.93365675,
  'label': 'bird_crow',
  'y_min': 279,
  'x_min': 640,
  'y_max': 340,
  'x_max': 767
  }]
}

The list of currently trained for objects in in birds_classes.txt

trainTest results are in train/trainTest.results.txt

debug.pics contains training run debug images highlighting where the model did not fully align with expected results, suggested map changes and such.

For more details on training see DeepStack training

For more help setting and or debugging a DeepStack setup see Quick Blue Iris with DeepStack debug

See runTrain.bat for an example of running training locally with my DeepStack Utils repo style setup.

Training results

confusion_matrix.png labels.jpg labels_correlogram.jpg precision_recall_curve.png results.png test_batch0_labels.jpg test_batch0_pred.jpg test_batch1_labels.jpg test_batch1_pred.jpg test_batch2_labels.jpg test_batch2_pred.jpg train_batch0.jpg train_batch1.jpg train_batch2.jpg